Query Fan-Out Tools Compared (2026): Simulators vs Recorders

TL;DR

Query fan-out tools come in two classes, and the class decides what you can report. Simulators such as Qforia, Wellows, OtterlyAI's free tool, Indexly and Rankability predict the sub-queries an engine would likely run; only Qforia and Wellows name a model behind the prediction (Qforia uses Gemini; Wellows cites GPT-4o as an example). Recorders such as Profound, Goodie, Radarkit, Peec AI and Ahrefs Brand Radar log the searches an engine actually ran in answers the tool triggered. Plan with a simulator. Report with a recorder. Recorders log the tool's own runs and the session extensions log yours, and no tool page we read says it shows the fan-out searches inside your buyers' own sessions. Google AI Overviews exposed no queries in 150 answers we pulled from Peec AI in October 2026.

Query fan-out tools compared by type (2026)

Query fan-out is Google's name for an AI system "issuing multiple related searches across subtopics and data sources" before it writes one answer. New to the concept? Start with what fan-out is. Search Engine Land: "These sub-queries are not shown to the user." Every query fan-out tool in the table is an attempt to get around that.

How we compared the tools

We classed each tool with one test, applied to that tool's own pages. If the vendor says it predicts, generates or simulates sub-queries, it is a simulator. If it says it captures or records the queries an engine ran inside real AI answers, it is a recorder. Prices come from each vendor's pricing page on 18 October 2026, read from a US connection.

What three AI engines searched: 450 answers, 15 days

In 450 AI answers LoudFace pulled from Peec AI exports between 4 and 18 October 2026, ChatGPT exposed its searches in 101 of 150 answers, Perplexity in 4 of 150 and Google AI Overviews in 0 of 150.

Across 10 buyer questions asked once a day for 15 days (4 to 18 October 2026), ChatGPT ran 326 distinct searches (counted per question), and 307 of them appeared only once.

The questions were the agency-buyer questions LoudFace tracks, the kind a B2B marketing lead asks before hiring help with SEO and AI search. Each engine answered every question once a day: 10 questions, 15 runs, 3 engines, 150 answers per engine. Every answer is a scheduled run Peec triggered, so this is category data about engine behaviour. Nobody's private session is in it.

ChatGPT never ran the same set of searches twice in a row. We compared 91 pairs of consecutive runs of the same question, counting only runs that showed queries. In 0 of the 91 did ChatGPT run an identical set. Mean Jaccard overlap was 0.018 and the median was 0 (Jaccard is shared queries divided by all distinct queries across the two runs, so 1 means identical sets).

On 5 October, asked about the best GEO agencies, ChatGPT searched best Generative Engine Optimization GEO agencies 2026, plus an exact-phrase GEO agency query. On 6 October the same question produced best GEO agencies Generative Engine Optimization agencies 2026, and a query naming Profound, Goodie and Omnius. No query string repeated.

The strings churn, but the habits hold. 80 of 351 ChatGPT queries used a site: operator. The most frequent targets were agencies' own domains: site:omniscientdigital.com (11 times), site:directiveconsulting.com (8) and site:skale.so (7). ChatGPT checks named agencies on their own domains. 88 of 351 queries carried the year 2026, and 0 simply restated the prompt. Outside research points the same way, at a higher rate: Konitzny's analysis, as summarised in Peec's docs, found site: in "roughly 43% of query fan-outs" after GPT-5.6, and "84% of targeted domains are branded".

LLMrefs says "Perplexity shows its fan-out process more transparently than other AI search engines." In our sample it showed queries in 4 answers out of 150. Visible sources are not visible queries.

Google AI Overviews showed nothing, 0 of 150. A recorder pointed at AI Overviews had nothing to record in this sample.

Simulators: tools that predict the sub-queries

A simulator predicts the searches an engine would probably run for your seed prompt. Some name the model doing the predicting (Qforia uses Gemini, Wellows cites GPT-4o as an example), and most do not. No engine run is observed. That is fine for planning and useless as proof. Semrush ran two simulators in its own experiment because "Different tools often surface different queries."

Qforia (iPullRank)

iPullRank describes Qforia as "a tool developed by Mike King at iPullRank to simulate a query fan-out." Per iPullRank, Qforia runs on "the same Gemini model used by AI Overviews and AI Mode, so the output is still probabilistic." You choose AI Mode or AI Overview mode, which is how Semrush set it up. The tool is free, but you must "enter a paid Gemini API key" first.

Best for: Google-first content planning. One limitation: nothing runs until you add a key, and the Gemini billing is yours.

OtterlyAI Query Fan Out

Otterly's free tool promises to "See the sub-queries ChatGPT and Google AI Mode generate from one prompt". The same page states the method: "By simulating the fan-out process, you can see which variations Google and ChatGPT might generate." Paid plans start at $29/month for Lite, with 15 search prompts.

Best for: a quick ChatGPT versus Google read on one prompt. One limitation: it predicts, so never report it as what ChatGPT searched.

Wellows Query Fan-Out Generator

Wellows will "Transform a single keyword into multiple semantically rich search queries." and describes its generator as "modern generative models (e.g., GPT-4o)". Paid plans start at $29 per month (Basic), priced in AI answers.

Best for: keyword researchers who want a clean CSV. One limitation: no target engine is named, so you get a general-purpose model's guess at another engine's behaviour.

Indexly Query Fan Out Generator

Indexly says it "expands your keyword into hundreds of semantically related, long-tail queries." You can "Download the entire dataset as a CSV file". Paid plans start at $99 per month (Starter).

Best for: mapping a big topic. One limitation: it names no target engine, and cutting hundreds of predicted variants down to a page plan is your job.

LLMClicks Query Fan-Out Coverage

LLMClicks offers to "Simulate the 20+ sub-queries Google AI generates and instantly check if your content covers them." It scores your page against that list. Pricing is a lifetime deal from $159 (Starter), and "All plans include a free 14-day trial."

Best for: checking one existing page before a refresh. One limitation: the sub-queries it simulates are the ones it says Google AI generates, and the coverage score grades you against predictions.

Rankability AI Search Query Fan Out

Rankability's input box sets expectations: "Type any question or topic to see how AI engines might expand and interpret it". It labels one output group with "These look like People Also Ask style questions." The tool is free with PNG export. Paid plans start at $97/month (Starter), and the trial needs a card.

Best for: a visual brief you can hand a writer. One limitation: an image export is hard to diff or track over time.

Relve Query Fan-Out and Grounding Checker

Relve generates 40 queries free, "then check grounding on the ones that matter most". The grounding step is live: "Each query is put to a live AI engine 3 times with a web search tool available but optional." It records whether the engine searched. It does not list the sub-queries. Relve Pro is $29/mo, priced in research runs; the pricing page does not say the fan-out tool is part of it.

Best for: finding which questions push an engine onto the web at all. One limitation: the sub-queries themselves are generated, and the live engine is not named.

queryfanout.io (New Chemistry)

The headline suggests you see what LLMs actually search. The method text says otherwise: "We generate the categorized sub-queries, then run live web searches on each to surface real cited pages." It is "Free, no credit card required", with the full report behind an email, and "We pull volume, KD, CPC, and AI Overview presence from real search data". No price is published.

Best for: predicted sub-queries with search data attached. One limitation: we class it as a simulator on its own method text, whatever the headline implies.

Recorders: tools that log the queries engines ran

A recorder sends prompts to a live engine on a schedule, through the chat interface or an API, and logs the searches the engine exposed. The sample is the runs the tool triggered. Some vendors describe a wider source: Goodie says it "monitors real customer queries across all major AI platforms", and Ahrefs says "Every prompt is modeled from real user searches". Neither page says it shows the fan-out searches inside a particular buyer's own session. For the wider tracking market beyond fan-outs, see our AEO tools comparison.

A recorder is only as good as the engine's openness. OpenAI's API docs say the web search action "usually (but not always) includes the search queries which were searched." The Gemini API response carries a google_search_call step that "Contains the search queries the model executed."

Profound Query Fanouts

Profound launched Query Fanouts on 8 October 2025 and says that for the prompts it sends, "we also capture the resulting query fanouts". Its trial "includes daily analysis of 50 unique prompts for 7 days across ChatGPT, Gemini, and Google AI Overviews", and ongoing tracking sits on Enterprise with no published price.

Best for: enterprise teams reporting fan-outs across several engines. One limitation: no public price, and the trial lists Google AI Overviews, which exposed no queries in the Peec sample. Ask what Profound shows there.

Goodie

Goodie's own guide says its platform "automatically captures fan-outs from millions of daily prompts across ChatGPT, Gemini, and more". The pricing page lists Core at $295 / month with a 7-day free trial. The same company's blog says the platform starts at $495 per month. We went with the pricing page.

Best for: teams that want prompt research and tracking under one login. One limitation: two prices on two of its own pages. Get the number in writing.

Radarkit

Radarkit says: "We open ChatGPT and other AI chat interfaces like a real customer, ask your prompts, and record every answer". It names "Google AI Overview, ChatGPT, Google AI Mode, Copilot, Gemini, and Perplexity", prompted through real chat interfaces rather than APIs. Lite starts at $29 /mo, and "Every plan starts with a 7-day free trial".

Best for: six named engines at a $29 entry price. One limitation: a real chat interface is still Radarkit's own session and tells you nothing about your buyer's.

Peec AI

Peec's docs define it this way: "A query fan-out is the set of hidden searches an AI engine runs behind one prompt." Its pricing page lists a Fanout overview among plan features and counts answers this way: "Calculated by multiplying tracked prompts by active models and tracking frequency." Plan prices are not in the page text.

LoudFace uses Peec AI to track AI answers for clients, and the 450-answer sample comes from its exports.

Best for: daily tracking of a fixed prompt set with fan-out views per model. One limitation: you can't read a price off the page, and Google AI Overviews returned 0 queries in our 150 answers.

Ahrefs Brand Radar

Ahrefs says Brand Radar's AI responses report shows fan-out queries for ChatGPT and Perplexity prompts. Custom prompts let you "Discover the fanout queries AI generates behind the scenes when processing a prompt to find content gaps." They start at €47/mo as served to a US visitor on 18 October 2026.

Best for: Ahrefs users who want fan-outs next to keyword data. One limitation: Brand Radar tracks prompts on seven platforms, but Ahrefs' own guide shows fan-out queries for ChatGPT and Perplexity prompts. Ask which platforms carry fan-outs before you buy for them.

Semrush Enterprise AIO

Semrush added Query Fan-Out Analysis to Enterprise on 15 January 2026. Semrush says it "reveals the Google search queries that fuel AI responses in systems such as ChatGPT", showing "what the fan-out queries were for the key prompts being tracked". It "is designed for enterprise organizations or growing teams", with no public price.

Best for: teams already on Semrush Enterprise. One limitation: the self-serve AI toolkit, listed at $99 per month, does not mention fan-out on its help page. Don't buy it for this.

Hybrids and session recorders

LLMrefs

LLMrefs does both jobs and labels them. The free tool says "Generate fan-out queries for user prompts with our QFO simulator." and "Choose which AI search engine to simulate." Then it tells you the limit itself: "But to see the actual fan-out queries generated by real AI search engines, you need AI visibility monitoring". The paid plan is $79/month and lets you "Track 500 prompts for brand mentions, sources and fan-out queries", with CSV export and API access.

Best for: starting free and moving to tracking without switching vendors. One limitation: simulated and recorded queries live in one account, so label which is which in every report.

Chrome extensions that record your own ChatGPT session

The ChatGPT Query Fan-Out Tool is "Built by Backlinko, a Semrush Company." Its store page says "The extension captures the response in real time as ChatGPT generates it", including "each sub-query the model ran". You can "Export any conversation as CSV", and a batch loader lets you "Enter multiple prompts (one per line) and run them against ChatGPT automatically." LLMrefs has a free sibling: "Reveal the search queries sent to Bing by ChatGPT web search with the LLMrefs Chrome extension."

Best for: recorded queries for your own prompts. One limitation: it records the session in your own browser, so it shows your searches and not your buyers'.

Dejan fan-out tool

Dejan sits between the classes. Its page says "Enter a prompt and see the fan-out a grounded model runs." The engine is not named, and Goodie's guide calls Dejan's output predictions rather than real ChatGPT or Gemini queries. It is free with a Google sign-in. Treat its output as one grounded model's run and never as ChatGPT's.

Best for: testing one prompt against a grounded model. One limitation: the engine is not named, so the output is not ChatGPT's or Gemini's.

What each type is good for, and where both stop

Simulators plan coverage

Use a simulator to list the sub-topics a page has to answer before you write it. Semrush did exactly that on four articles: citations rose from two to five, while share of voice fell from 23.4% to 20.0%. Even Semrush says there is no way to know which queries AI Mode will use, "since these are generated dynamically and can vary with each search".

So treat the output as a coverage checklist. Ahrefs warns that "Where many people go wrong is thinking that these queries are like topic clusters 2.0". Writing one page per predicted query is how you end up with a pile of thin pages.

Recorders measure, on engines that expose queries

A recorder earns its fee when you report, but read it as a pattern across many runs. With a Jaccard overlap of 0.018 between consecutive runs, one recorded set is close to noise. The signal is in the habits, such as site: checks on named firms and the 2026 modifier.

Vocabulary is where a recorder pays for itself. In LoudFace's tracking, ChatGPT searches for GEO and AI search optimization agencies, even though we say AEO. The October sample repeated the pattern, with GEO wording and the 2026 modifier in the searches. If your pages only say AEO, you are missing the words the engine types.

Recorders also stop at Google's wall. Google says sites appearing in AI features are "included in the overall search traffic in Search Console", and AI Overviews exposed 0 queries in 150 answers. For a Google read, measure share of answer on the prompts you care about instead of hunting for queries Google does not publish.

What the tools don't claim to show

No tool page we read says it shows the fan-out searches inside your buyers' own sessions. Goodie says it monitors real customer queries and Ahrefs models its prompts from real user searches, but neither page says its fan-outs come from a particular buyer's session. The extensions see only yours. And fan-out queries rarely have search volume. Ahrefs found that "over 95% receive no recurring searches", and Relve says "most conversational questions sit below every keyword tool floor". A volume column next to a fan-out query is mostly a column of zeros.

The third signal sits on your own server. Server logs show which pages AI bots actually fetched from your site, which neither class can tell you. Our server log playbook covers how to read them.

Which type you need

If anyone puts simulated queries in a report as "what ChatGPT searched", send it back. Predicted queries are a planning input. Recorded queries, read per engine across many runs, are evidence. If you want to know what the engines say about your brand today, start with an AI visibility audit.